Project Summary/Abstract Millions of people suffer from some form of paralysis. In most of these cases the connection between the brain and the spinal cord is damaged, however, the motor cortex is healthy and intact. Thus, for these individuals, brain-machine interfaces (BMIs) hold significant promise for improving quality of life. BMIs decode an individual's intention to move by utilizing statistical models of neural activity patterns recorded from the motor cortex using implanted electrode arrays. While these methods have been encouraging in preclinical experiments and clinical trials for controlling thought-driven 2D computer cursors, they suffer from poor performance when applied to higher degrees-of-freedom (e.g., robotic limbs), and are not robust to the inevitable degradation of the electrode array. In order to address these clinical needs, this project starts from the recent observation that just as some behaviors are easier to learn, some patterns of neural activity, termed neural states, are also easier to generate. The overarching goal of this project is to elucidate if these ?easy to generate? neural states can be used to robustly control a prosthetic arm. This is a significant departure from current decoding methods, which incorporate little to no information about the motor system, especially its ability to learn and adapt. The first major aim of this work is to develop experiments and analysis methods in order to find these ?easy to generate? neural states in the non- human primate (i.e., rhesus monkey) motor system. Here ?easy to generate? can be understood as the monkey's ability to volitionally generate that particular neural state. The second major aim of this work is to characterize the properties of the motor system that enable some states to be more easily generated than others. Prior work in our lab has shown that motor cortical population activity has well-defined structure, as predicted by dynamical system theory. These dynamics cause neural states to evolve in lawful ways through time. The work here will extend these findings by characterizing the dynamics associated with a monkey learning to generate a neural state. Finally, the third major aim of this work is to determine if neural states that monkeys can volitionally generate can be utilized for robust control of a prosthetic arm. The central hypothesis of this work is that building a model that only utilizes firing patterns that can be easily generated (as determined experimentally) will enable robust and high-performance control of a prosthetic arm. If successful, this study could have significant clinical impact by presenting a new paradigm to enable robust control of a prosthesis.